Qualitative traffic analysis using image processing and time-delayed neural network

We present an online, feature-based approach to estimate traffic qualitative parameters from a sequence of traffic images. Considering the factor of time and attempting to simulate human behavior, a time-delay neural network is used to determine the traffic status through traffic lanes. The acquired frames are divided into a number of blocks based on number of lanes and road boundary coordinates, which are obtained automatically by a part of the system called the road boundary detection system. Two extracted principal features from each block of a lane which are vehicle detector and movement detector will form the input vector of the neural network. The neural network classifies each lane into a level of traffic congestion. The neural network was previously trained with various traffic and different lighting conditions. Finally a description of traffic scene is obtained using descriptions of all lanes.